Developmental noise and ecological opportunity across ... · plasticity and environment‐dependent developmental in-stability. This paper does not attempt to characterize the behavior

Evolution & Development. 2019;e12305. wileyonlinelibrary.com/journal/ede © 2019 Wiley Periodicals, Inc. | 1 of 12https://doi.org/10.1111/ede.12305

DOI: 10.1111/ede.12305

RE S EARCH

Developmental noise and ecological opportunity acrossspace can release constraints on the evolution of plasticity

Jeremy Draghi1,2,3

1Department of Biology, Brooklyn CollegeCUNY, Brooklyn, New York2The Graduate Center, CUNY, New York,New York3Department of Biological Sciences,Virginia Tech, Blacksburg, Virginia

CorrespondenceJeremy Draghi, Department of BiologicalSciences, Virginia Tech,Blacksburg 24061, VA.Email: [email protected]

Funding informationNSF, Grant/Award Number: 1714550

Abstract

Phenotypic plasticity is a potentially definitive solution to environment

heterogeneity, driving biologists to understand why it is not ubiquitous in

nature. While costs and constraints may limit the success of plasticity, we are

still far from a complete theory of when these limitations actually proscribe

adaptive plasticity. Here I use a simple model of plasticity incorporating

developmental noise to explore the competitive and evolutionary relationships

of specialist and generalist genotypes spreading across a heterogeneous

landscape. Results show that plasticity can arise in the context of specialism,

preadapting genotypes to later evolve toward plastic generalism. Developmental

noise helps a mutant with imperfect plasticity successfully compete against its

ancestor, providing an evolutionary path by which subsequent mutations can

refine plasticity toward its optimum. These results address how the complex

selection pressures across a heterogeneous environment can help evolution find

paths around constraints arising from developmental mechanisms.

1 | INTRODUCTION

Phenotypic plasticity is the development or modification oftraits in response to environmental cues. Adaptive plasticityis common but not ubiquitous (Palacio‐López, Beckage,Scheiner, & Molofsky, 2015), and biologists feel the lack of apredictive theory for when plasticity should evolve whenthey contemplate the limits to species’ resilience to environ-mental change (e.g., Ashander, Chevin, & Baskett, 2016).Plasticity represents both a product of adaption and aprecursor to further change. Plasticity is one solution toenvironmental heterogeneity, and a number of theorists havepushed toward a synthetic theory that explains whenplasticity will emerge and preclude other results likegenetically differentiated local adaptation, stochastic bet‐hedging, or simply extinction (e.g., Bull, 1987; Scheiner,2014b; Svardal, Rueffler, & Hermisson, 2011; Tufto, 2015).Plasticity has also been linked to speciation (Pfenniget al., 2010; Schneider & Meyer, 2017), innovation (Moczek

et al., 2011), and invasion success (Davidson, Jennions, &Nicotra, 2011), placing it at the core of an emerging,predictive theory of the determinants of the rate of evolution.

Understanding the limits of plasticity is a persistentchallenge in evolutionary biology because the issuerequires substantial attention to ecology, life‐history,and development. Ecology provides both the stick and thecarrot, providing both the need for distinct phenotypesacross a heterogeneous environment and the informativecues necessary to produce the correct phenotypes. Life‐histories can limit the reliability of this informationbecause movement preceding reproduction or metamor-phosis can divorce past cues from the organism’s adultenvironment (Scheiner, 2013). The limitations imposedby dispersal from the natal environment depend on thelags and costs intrinsic to the developmental system:behaviors might adjust rapidly to changes in cues, whilemorphology might be less malleable. Efforts to review thefield have produced taxonomies of the various costs that

reduce the realized benefit of plasticity and theconstraints that prevent it from arising at all (DeWitt,Sih & Wilson, 1998; Murren et al., 2015). While somecosts are obvious—for example, the need to produce asensory organ—comparative work has mostly failed tofind clear costs that make plasticity less competitive thanclosely related nonplastic specialists (Auld, Agrawal, &Relyea, 2009; Van Buskirk & Steiner, 2009). Attentionhas, therefore, shifted to understanding how constraintsmight restrict the evolution of plasticity to certain traits,taxa, and environments.

The complex and poorly understood effects on theevolution of plasticity stemming from these diverseecological and organismal factors motivate theoreticalresearch. Exploration of extremely simple models canprovide structured hypotheses for where constraints mightcome from and how experiments can be designed tomeasure which constraints actually matter for explainingthe distribution of plasticity across nature. The modelpresented here stems from a tradition in which modeling ofplastic development is made extremely simple in favor offocusing attention on complexity arising from environmentsand life‐histories. Here, an organism’s genotype determinesthe slope and intercept terms of a linear equation, withplasticity emerging as the product of the slope and thedifference in cues provided by each environment. Whileabstract, similar approaches have been successfully used tomodel the interactions of spatial and temporal heterogene-ity with life‐histories (Scheiner, 2013), the joint determina-tion of dimorphism by plasticity and genetic differentiation(Leimar, Hammerstein, & Van Dooren, 2006), and themaintenance of genetic variation in plastic traits (De Jong &Gavrilets, 2000). A slope‐intercept model codifies the ideathat the effects of mutations are biased by some develop-mental structure, but that this structure can itself evolve tochange those biases. The work here combines this simpleframework with both spatial structure and developmentalinstability to articulate new hypotheses about the ecologicaland developmental factors favoring plasticity.

Developmental instability has been studied withinevolutionary biology under names like canalization, robust-ness, and specific cases such as fluctuating asymmetry. Arecent surge of interest in the topic has been inspired by thetremendous growth in capacity to measure and modelstochastic heterogeneity at the levels of molecules, cells, andfitnesses in single‐cell organisms (Bruggeman & Teusink,2018; Draghi, 2018; Raser & O’shea, 2005). Stochasticheterogeneity in phenotypes arising from developmentalinstability has often been viewed as both a direct detrimentto fitness and an impediment to the fixation of adaptivemutations (Wang & Zhang, 2011). However, the concept ofrandom phenotypic diversity as an adaption to unpredict-able change has a long history in evolutionary thought

(Starrfelt & Kokko, 2012). Microbial experiments have beenparticularly influential by illustrating how cells can usestochastic variability to create a diverse portfolio ofphenotypes—e.g., nongrowing, resistant spores and cellscompetent for natural genetic transformation (Veening,Smits, & Kuipers, 2008). While most attention has focusedon understanding how this stochastic variability couldfunction as a form of bet‐hedging (Starrfelt & Kokko, 2012),recent experiments point to more subtle benefits tovariability and highlight its potential role in stimulating,rather than slowing, adaptation (Bódi et al., 2017).

Although plasticity and developmental instabilityhave often been considered separately, a recent findingof a positive correlation between these traits in Arabi-dopsis phenotypes (Tonsor, Elnaccash, & Scheiner, 2013)has helped to spark interest in their joint consideration.A recent modeling paper showed how instability arisingfrom genetic factors could evolve, and that evolvedrobustness in development actually constrained theevolution of plasticity (Draghi, in review). The coreresult of this paper was that developmental noiseloosened pleiotropic constraints preventing plastic mu-tants from successfully competing against their parentgenotypes, allowing a population to evolve a plasticresponse to environmental heterogeneity. This priorwork was confined to the limitations of most traditionalpopulation‐genetics models: a single well‐mixed popula-tion reproducing under soft selection. Also, stochasticheterogeneity arose solely from genetic factors in thisprior work, making it difficult to isolate the effects of thattrait from other traits determined by the same geneticfactors. Here I expand upon this study by applying hardselection to a population arrayed across a heterogeneouslandscape, and by allowing the environment, rather thanthe genotype, to control the expression of developmentalnoise. The model explored here allows plasticity to arisein response to large‐scale differences between distinctenvironments, in the context of stochastic inputs todevelopment that also derive from the environment.While this dual role of the environment on developmentconnects with classical ideas about plasticity (Bradshaw,1965), its formulation in this spatial model allows for newinsights about the origins of genotypes using plasticity toachieve a generalist phenotype. Specifically, the ability ofa population to adapt to a novel environment depends onboth the aid of developmental noise and the prioremergence of genotypes using plasticity to achieve novelspecialist lifestyles. These results show how the develop-mental biases intrinsic to a very simple form of pleiotropycreate a soft constraint on the evolution of an innovativefeature, and illustrate how environmental opportunityand evolutionary history can lead to lineages thatovercome that constraint.

2 of 12 | DRAGHI

2 | MODEL AND METHODS

The purpose of this model is to explore the interplay of avery simple developmental system with a more complexecological framework featuring hard selection, explicitspatial structure, and two types of environmental effects:plasticity and environment‐dependent developmental in-stability. This paper does not attempt to characterize thebehavior of this complex model over the range ofparameters; rather, the goal is to generate hypotheses aboutthe role of space in the evolution of plasticity, and toillustrate how a developmental system leads to a biased setof mutations that interact in subtle ways with selectionacross a landscape. To support these goals, I will aim toarticulate the model I used as simply as possible, ratherthan present it in a more general notation.

The spatial environment for a population was modeled asa lattice of cells, each of which could contain up to oneorganism. These cells were arranged in a rectangle of widthW and height H, allowing a maximum of N=W×Horganisms at a time. Each cell was assigned to one of twoenvironments, labeled 1 and 2: these environments differedin their optimal value of the organism’s trait, labeled zopt(1)and zopt(2), respectively. The environment also determinedwhich of two values of a cue would be perceived by anorganism developing in that cell, providing the developingorganism with perfect information about the optimalphenotype for that environment.

Each cell in a landscape was assigned to eitherEnvironment 1 or 2 based on a sigmoid function of the xcoordinate of that cell’s location. Environmental noise wasintroduced during development into the phenotype, with amagnitude increasing with the y coordinate. This allowed fortwo forms of stochasticity: dispersal introduced a chancecomponent to where an organism would develop, and

environment noise within an environment shaped theresulting adult phenotype.

The probability that a cell would be assigned toEnvironment 2 was determined as follows:

⎜ ⎟⎛⎝

⎞⎠( )p x e( ) = 1 + .

xW

2−2 2 −1

−1

(1)

A clustering algorithm was then used to produce avariable degree of autocorrelation in space whilerespecting the expected frequencies derived fromEquation (1). This algorithm chose a pair of cells withthe same x coordinate and inspected their assignedenvironments and the eight cells making up theirimmediate neighbors. If the chosen cells were assigneddifferent environments, and if each was environmen-tally dissimilar to the majority of its neighbors, then theassignments for those two cells were swapped; other-wise, no change was made. The number of random pairsconsidered for a potential swap was determined as aPoisson random number with a mean of θN, where θacts as a clustering parameter. Figures 1a and 2 showexamples of the resulting spatial structure when θ= 1.

An organism’s genotype consisted of two real numberscorresponding to the slope, a and intercept, b of thecanonical equation of a line. The phenotype correspondingto an organism’s genotype was determined from threesources: these two genotypic parameters, the cue c(i)associated with the environment assigned to that particularorganism’s location, and a Gaussian noise term with amean of zero and a standard deviation σ(y) = 50+ 300(y/H− 1). A genotype’s phenotype is, therefore, a randomvariable described by the following equation.

z x y ac i b N σ y( , ) = ( ) + + (0, ( )). (2)

FIGURE 1 (a) A portion of an example landscape showing the distribution of patches of the two environments across the dimensions ofthe space. The clustering parameter was θ= 1 for this example. The color of each cell indicates the Environment (1 or 2), with the frequencyof Environment 2 increasing from left to right. Environmental noise causing developmental instability increases from bottom to top (notillustrated). (b) An illustration of a plastic mutant (phenotypic distributions in light gray) derived from a static ancestor (dark gray). Boldlines indicate the fitness functions in the two environments. Environmental noise is σ= 150 for the lower examples and σ= 250 for those ontop, representing the change in environment noise along the y‐axis of the landscape. For the ancestor, the genotypic parameters are a= 0and b= 1,000 (see Equation (2)); for the mutant, a= 0.3 and b= 1000 [Color figure can be viewed at wileyonlinelibrary.com]

DRAGHI | 3 of 12

The distribution of phenotypes produced by a givengenotype, therefore, depends on environmentalfactors in two ways. A genotype with some degree ofplasticity (a ≠ 0) will produce a different mean pheno-type in Environment 1 compared to its average inEnvironment 2 (Figure 1b). In addition, the y coordi-nate shapes the variability of development but not itsmean; this stochastic influence occurs regardless of thetype of environment or the organism’s genetic values.

An organism’s fitness is assigned via a Gaussianfunction comparing its phenotype with the optimum forits environment i. Throughout the simulations reportedhere, z =1000opt,1 , zopt,2= 2000, c1 = 1000, c2 = 2000, andσ =5,000opt2 . Fitness w is therefore:

⎛⎝⎜⎜

⎞⎠⎟⎟w

z z

σ= exp

−( − )

2.

iopt,2

opt2

(3)

After birth, an organism disperses based on aGaussian movement kernel with a variance of 2.5. Ifthe organism disperses to an occupied cell or off themargins of the landscape it is lost from the population;if it lands in an empty cell it develops to adulthood asgoverned by its genotype (Equation (2)). Fitnessdetermines fecundity and reproduction is semelparousand asynchronous. A generation is defined by permut-ing the list of all cells and processing each cell in orderaccording to the following algorithm: if a cell isoccupied, that organism produces a Poisson‐distributednumber of offspring with a mean equal to 10 times thefitness of the organism; these offspring then disperse,and the focal organism is removed.

Organisms are haploid and reproduce asexuallywith mutation at a rate μ = 0.001/individual/genera-tion. This rate was chosen to balance two constraints: avery high rate can allow a lineage to acquire two ormutations in rapid succession, allowing pleiotropic

constraints to be avoided with unrealistic ease. A ratethat is too low simply slows the evolutionary dynamicsand inflates the needed computation time. By cappingthe rate at an average of one mutation per 1,000generations we err toward the side of computationalinefficacy and allow selection to act effectively onindividual mutations.

A mutation affects either the a or b parameter withequal probability. Mutations in the a parameter areGaussian with a standard deviation of 0.4; mutations inthe b parameter are also Gaussian with a standarddeviation of 500. Because genotypes are fully linked,each individual has a single lineage of ancestors.Genealogies of every new mutant genotype arerecorded during each simulation such that lines ofdescent can be unambiguously reconstructed andtraced. The spatial location on the landscape at whicha new mutation originates is also recorded.

We can write a useful equation for the relativefitness of a genotype in a given environment, averagingover the distribution of phenotypes it would beexpected to produce. The integral of Equation (3) withthe Gaussian distribution of phenotypes defined byEquation (2) is given below:

⎛⎝⎜⎜

⎞⎠⎟⎟

⎛⎝⎜

⎞⎠⎟

⎛⎝⎜⎜

⎞⎠⎟⎟

∫∞

∞ω μ σ

σ π

z z

σ

z μ

σdz

ω μ σμ μ

σ σ

σ

σ σ

( , ) =1

2exp −

( − (1))

exp −( − )

,

( , ) = exp −( − )

2( + ) +.

2

−

opt2

opt2

2

2

2 opt2

opt2 2

opt

opt2 2

(4)

Simulations and analysis scripts were written inR and will be made available in a Data Dryad repository.

FIGURE 2 Colonization of anexample landscape by the ancestralEnvironment‐1 specialist. Snapshot of theadults in the population after 1,000generations, showing a quasiequilibriumdistribution without evolution [Colorfigure can be viewed atwileyonlinelibrary.com]

4 of 12 | DRAGHI

3 | RESULTS

3.1 | Model dynamics without evolution

Simulations began with ancestral organisms thatlacked plasticity and were perfectly adapted to one ofthe two environments, referred to as Environment 1 orthe ancestral environment. Moving east across thelandscape the frequency of patches of this ancestralenvironment declines (Figure 1a) in favor of Environ-ment 2, the novel environment, to which the originalgenotype was poorly adapted. Figure 2 shows anexample of the spread of the ancestral genotype in asimulation without the potential for evolution (mutationrate μ= 0). As in all the simulations discussed here,individuals were initially placed on the western edgeof the landscape and moved across it via dispersalof offspring. This distribution represents a quasi‐steady‐state, as some of the unoccupied clusters of Environ-ment 1 might be colonized by rare long‐distancemigrants given sufficient time. However, this snapshotdoes illustrate key aspects of the spatial distribution ofthe original specialist. Notably, developmental stochas-ticity reduces fitness (see Equation (4)) which, underthese hard‐selection conditions, causes the populationdensity to decline along a latitudinal cline. Thecombination of the latitudinal gradient of environmentnoise magnitude with the lateral gradient in thefrequency of Environment 2, to which this genotype ispoorly adapted, creates a roughly diagonal range marginfor this genotype. Finally, note that autocorrelationamong the patches of each environment leads to smallenclaves of the population in areas surrounded by theunfavorable Environment 2, as well as numerousuncolonized patches of Environment 1.

3.2 | Evolution of model generalists andspecialists across space

Evolutionary simulations with mutations in bothgenetic parameters produced several distinct phenotypes.Figure 3 shows the origination points of successfulmutants of two types: specialists on the novel environ-ment, and plastic generalists that use both environments.A genotype whose fitness in Environment 2 is at leastfour times greater than its fitness in Environment 1 isconsidered a specialist on Environment 2; specialists onEnvironment 1 are defined similarly, and generalists arethen defined as the intermediate cases. These mutantsare drawn from 100 replicate simulations, and a mutantwas classified as successful if it rose to at least 2% of itspopulation—qualitative patterns were not sensitive to thevalue of this threshold (data not shown). Both types ofmutants can thrive across a range of origins in thex coordinate, though both show some clustering towardthe right half of the middle. However, in the y coordinatenew specialists tend to arise in areas with lowerdevelopmental noise, while plastic generalists showan opposite trend and overall show more variability inwhere they flourish.

The role of environmental noise in the origination ofsuccessful plastic generalists can be understood by firstappreciating how noise shapes both the fitness and therealized resource utilization of a genotype. The mutantsin Figure 3 are assigned to the categories of generalistsand specialists based on their predicted niche, given thelevel of environmental noise present where each mutantarose. While the basic pattern in Figure 3 is robust to thearbitrary fourfold cutoff applied (data not shown), therole of environmental noise in this determination isintrinsic to the model and deserving of more explication.

FIGURE 3 Novel specialists and plastic generalists arise from different areas of a heterogeneous landscape. The figure comprises theresults of 100 replicate evolutionary simulations; points depict successful mutants that specialized in Environment 2 or showed a significant,plastic ability to use both environments. Mutants were defined as successful if they achieved a maximum frequency of at least 0.02 duringthe simulation (see Section 2) [Color figure can be viewed at wileyonlinelibrary.com]

DRAGHI | 5 of 12

Therefore, Figure 4 illustrates how two generalistmutants compare with their ancestors across the rangeof environmental noise encountered in the simulatedlandscape. Changing the slope parameter a in thenonplastic ancestor can improve fitness in Environment2 but produces maladaptation in Environment 1; whendeveloped with greater environmental noise, the benefitof this mutation is improved and the cost lessened, bothin absolute terms and in comparison to the ancestor(Figure 4b). This mutation would, therefore, be deleter-ious in low‐noise environments and neutral or beneficialat higher levels of developmental noise. Another patternvisible in this example is that the nonplastic ancestor isinvariably much more fit in Environment 1 than inEnvironment 2, and the plastic mutant is always arelative generalist, with about equal potential perfor-mance in each environment. A second example slightlycomplicates this picture: a highly plastic mutant canfunction as a specialist for the novel Environment 2 atlow levels of noise, but display more equitable fitnessacross both environments at very high levels of noise(Figure 4c,d). This example also illustrates a hypotheticalpathway toward the evolution of an adaptively plastic

generalist: a plastic mutant might initially function as aspecialist in Environment 2, then serve as an ancestor fora generalist with equal plasticity (equal a) but moreequitable performance across environments.

The examples in Figure 4 motivate caution inattempting to infer the niche of a model organism basedsolely on its genotypic parameters. Specifically, aspecialist in a low‐noise environment can behave as ageneralist in a high‐noise context, and plasticity can beadaptive by producing a specialist on the novel environ-ment, rather than a generalist. Bearing these caveats inmind, we can use these niche predictions to formulatehypotheses about how environments interact with thedevelopmental system to facilitate the evolution ofplasticity. Figure 5 shows two example populations inwhich, according to these predictions, generalism aroseand came to exclude both the ancestor and evolvedspecialists for Environment 2. In both examples, theultimate descendants show slopes approaching one andintercepts approaching zero, which should produceoptimal plasticity at any level of environment noise. Ineach case, these highly plastic genotypes result from anumber of refining mutations, forming a chain of

FIGURE 4 (a) Illustration of the reaction norms of the ancestral specialist (dotted line) and a potential mutant in which the a

parameter has increased from 0 to 0.3. Gray lines show the fitness function in each environment. (b) Fitness in each environment for bothmutants over the range of values of environmental noise. Values were calculated based using Equation (4) with σopt

2 = 5,000. (c) As in (a) but

the Environment‐2 specialist has parameters a= 0.5 and b= 1,000, and the mutant changes the value of b to 700. (d) Fitness in eachenvironment for each genotype shown in (c)

6 of 12 | DRAGHI

ancestors that are predicted to derive fitness from bothenvironments. However, the predecessors to these plasticgeneralists are more varied and suggest a complexdynamic. In both examples, a plastic genotype predictedto be an Environment‐2 specialist is ancestral; in oneexample, a derived Environment‐1 specialist forms abridge between this ancestor and the first plasticgeneralist.

3.3 | Developmental and ecologicalprerequisites for the evolution of plasticgeneralists

To investigate the evolutionary processes suggested bythese examples, I first quantified how often evolvedspecialists on Environment 2 were plastic, as opposed tocarrying a mutation in their intercept parameters. Out of187 observed Environment‐2 specialists that achieveda maximum frequency of at least 0.02, 131 (70%)had changes only in the a parameter, indicating thatthey had evolved to become plastic. Of the remainder,

47 showed change in the intercept term but not the slope,while nine showed changes in both parameters. Specia-lists in the novel environment could, therefore, evolve inways that introduced either strong environmental sensi-tivity or maintained the ancestor’s insensitivity. These187 successful specialists were spread across all 100replicate populations, suggesting that distinct specialistgenotypes temporarily coexisted or arose serially in somereplicates. To examine the causal influence of specialistgenotypes on the evolution of plasticity I quantified howoften generalism evolved, and whether its evolution wasstrictly dependent on the presence of a plastic specialistancestor. Thirty‐nine of the 100 replicate populationsevolved to be dominated by generalists (i.e., thecombined proportion of predicted generalists exceeded0.5). In every case, the line of descent included an evolvedEnvironment‐2 specialist that was plastic, rather than aspecialist evolved via a mutant in the intercept para-meter. This supports a model in which the particulartype of novel‐environment specialist that happens toevolve in a population determines whether plasticity can

FIGURE 5 (a,b) Frequencies of predicted specialists and generalists over time for two example populations in which generalismevolved. Predicted niches are based on a genotype’s comparative fitnesses in each environment for the level of environment noise at which itevolved. (c,d) Genetic parameters for the genotypes along the line of descent of the most common genotype at generation 5,000 for the twopopulations in (a,b). Symbols indicate the predicted niche of each genotype; the origination time of each mutation is indicated by the tickmarks in (a) and (b)

DRAGHI | 7 of 12

readily emerge. I next evaluated whether the pattern seenin Figure 5a,c was general: did the line of descenttypically include a secondarily evolved specialist onEnvironment 1? This was the case for 82% (32/39) ofthe instances of successful generalism.

The finding that a secondarily evolved specialist onthe initial environment often served as a stepping‐stoneto generalism is explicable genetically by inspection ofFigure 3: a large decrease in the intercept term of a plasticEnvironment‐2 specialist can produce a genotype that itbest‐adapted to Environment 1 but poised to also exploitEnvironment 2 after a subsequent mutation increases theslope parameter. However, the emergence of suchgenotypes is ecologically puzzling because the ancestralgenotype is perfectly adapted to Environment 1 andwould presumably exclude less well‐adapted, secondarilyevolved competitors from that niche. One hypothesis thatmight resolve this paradox is that the actual niche of thissupposed specialist might be more general. A secondhypothesis is that these genotypes are not advantageous,even locally, but represent deleterious mutations thatnever achieve high frequencies, and are only found onthe line of descent because their enable a second,beneficial mutation conferring an adaptive generalistphenotype. Finally, inspection of Figure 2 suggests a thirdhypothesis: secondarily evolved specialists may be grow-ing within clusters of Environment 1 that are inaccessiblespatially to the bulk of the ancestral population.

To better understand these ecological dynamics ofcompetition among genotypes, I traced the populationsize and realized niche of each genotype on the line ofdescent of those populations that evolved a highfrequency of generalism. The realized niche, measuredas the proportion of a genotype’s reproductive outputderived from Environment 2, was calculated directlyfrom the fitnesses of all individuals of that genotype,rather than predicted as described above. Viewedthrough this lens, a genotype’s niche could change asit spread through the landscape and experiencedcompetition with other lineages. Figure 6 shows thepeak population size of each genotype that wasclassified as a secondarily evolved Environment‐1specialist, as well as realized niche of that genotypewhen it was most prevalent. It is evident that some ofthese putative specialists are actually deriving asubstantial proportion of their fitness from Environ-ment 2, and these genotypes also tend not to reachsubstantial numbers. However, other examples pro-duce substantial subpopulations primarily on patchesof Environment 1. Inspection of the spatial locations ofthese more abundant mutants suggested that theyprimarily arose near the range margins of the ancestralspecialist and spread along or expanded those margins.

However, the edge of a genotype’s range was suffi-ciently labile to make it difficult to quantify thisobservation.

Finally, I sought to clarify whether predicted generalistsdid, in fact, gain fitness by using both environments. Figure7 plots the realized niche measurements for each genotypeon the line of descent, using the two example populationsplotted in Figure 5. These examples are representative ofthe qualitative pattern seen in other populations that evolvegeneralism: successive generalist genotypes tend to favorone environment or the other and can coexist for moderateperiods of time alongside related generalists with differentbiases. Each line represents a single genotype, meaning thatvertical movement of a line represented ecological, notevolutionary change; specifically, a range shift or expansionthat changes the distribution of environments encounteredby that subpopulation. These figures illustrate that succes-sive refinement of the dominant generalist genotype occursin the context of dynamic repartitioning of the nichethrough both evolution and range shifts.

4 | DISCUSSION

A primary result here that extends and supports a findingfrom related previous work (Draghi, in review) showsthat developmental noise can actually aid the evolution ofplasticity under certain circumstances. While this pre-vious study demonstrated this argument in a highlysimplified ecological context, the results shown hereshow that noise can stimulate the evolution of plasticityeven when the fitness costs of noise reduce the localpopulation size in high‐noise environments. Althoughmany more organisms reside in low‐noise environment

FIGURE 6 Maximum subpopulation size of secondarilyevolved Environment‐1 specialists and their realized niche,defined as the proportion of their fitness derived from Environment2, at that peak

8 of 12 | DRAGHI

(Figure 2), and those environments are the most effectiveincubators of novel specialists (Figure 3), the associationbetween noise and the origination of plasticity is stillquite strong (Figure 3). A second, weaker pattern is thatsuccessful plastic mutants are more likely to arise inareas where the novel environment is quite common,again despite the paucity of organisms that live andreproduce in these areas. A general prediction emergingfrom these results is that variation in how selection actson new, innovative mutations may dominate over theinfluences of demography, perhaps making innovationsmore likely at a population’s periphery rather than in themore populous core.

While only a few other models of the evolution ofplasticity have focused on developmental noise orinstability, their conclusions and approaches have beenquite different (Scheiner, 2014b). One factor that mightaccount for this difference is that developmental noise isoften viewed as an inherent cost of plasticity (DeWitt et al.,1998; Scheiner, Caplan, & Lyman, 1991; Tonsor et al.,2013). This viewpoint is based on the idea that thesensitivity to the environment required for plasticity willnecessarily introduce more noise into development via anyrandomness or variation in the cue. One assumption ofthis argument is that noise is always disfavored becausethe resulting developmental instability produces amismatch between the genetically determined phenotypeand the optimum phenotype to which the population hasadapted. However, random variation in phenotypes can bebeneficial when environmental optima are difficult topredict, and this type of adaptive response has been

studied extensively under the umbrella of the term“diversifying bet‐hedging” (Frank & Slatkin, 1990; Starrfelt& Kokko, 2012). A number of recent papers havecombined these to examine how bet‐hedging, develop-mental instability, and phenotypic plasticity interact.Scheiner and Holt (2012) showed that extreme plasticitycould evolve as a form of bet‐hedging, and follow‐upmodeling work showed that developmental instability andplasticity could act as mutually exclusive strategies(Scheiner, 2014a, 2014b). Other recent approaches haveexamined how uncertainty in cues affects the relativevalue of bet‐hedging and plasticity (Donaldson‐Matasci,Bergstrom, & Lachmann, 2013) and empirical work hasbegun to disentangle how genotypes might vary in thedegree to which they use each strategy (Simons, 2014).Other approaches have modeled how a bet‐hedgingbenefit allows new regulatory connections to evolve evenin the absence of a correlation between that regulatorysignal and optimal phenotypes (Maxwell & Magwene,2017; Wolf, Silander, & van Nimwegen, 2015). These latterstudies provide a perspective on the relationship betweendevelopmental noise and plasticity that complements theresults here: in both, developmental instability provides aform of generalism by allowing at least some individuals toexploit several of the resources that they encounter.However, a key departure from these studies is that theresults here emphasize how developmental noise helps amutant with imperfect plasticity successfully competeagainst its ancestor, providing an evolutionary path bywhich subsequent mutations can produce a more refinedform of plastic generalism.

FIGURE 7 Realized niches, definedas the proportion of their fitness derivedfrom Environment 2, of the genotypes onthe line of descent of the dominantmembers of the final populations. (a) and(b) show the same populations as Figures5a,c and 5b,d, respectively. Widths of eachline are proportional to the square root ofthat genotype’s frequency, and serve as ascaled index of abundance of eachgenotype [Color figure can be viewed atwileyonlinelibrary.com]

DRAGHI | 9 of 12

Here I created a scenario in which the ancestralgenotype was well‐adapted to one environment butpoorly adapted to a second environment that wasconcentrated at the margins of its range. This set‐up,along with the strong limitation of linear dependencebetween cues and plastic responses, created a majorconstraint on the evolution of plasticity: any mutant thatchanges plasticity will have pleiotropic effects, and in theinitial genotype such a pleiotropic mutant must decreasefitness in the ancestral environment. The scenario of awell‐adapted population encountering a new environ-ment is found elsewhere in the literature on plasticity: forexample, Via (1987) examined this case in a quantitative‐genetics model that also included developmentalinstability. While environmental noise played a signifi-cant role in helping plastic generalists evolve in theresults presented here (Figure 3), the genotypes ofevolved specialists on the novel environment emergedas a critical ingredient in how evolution bypassedconstraints on plasticity. Plasticity was one mutationalpathway by which organisms could drastically changetheir phenotypes to match the optimum of the novelenvironment, and lineages that took this path tospecialism laid the foundations for the further evolutionof genotypes that applied that plasticity to a generalistniche (Figures 4, 5, and 7). Starting with a well‐adaptedspecialist reorients the question being answered by themodel: rather than generally addressing how plasticitycan evolve, the results speak more directly to howgeneralism can evolve via plasticity from a specialistancestor. The hypotheses generated by these results helpdemonstrate the potential value of taking this moreniche‐centric approach to familiar questions aboutconstraint and pleiotropy.

The dynamics of the evolution of plasticity as exploredin this model are linked to the genetic underpinningspecified for the plastic response: separate slope andintercept terms. While this decomposition of the deter-minants of plasticity has a long pedigree, it can hardly bemotivated biologically. Moreover, the decision is notwithout consequence: for example, constructing a linearreaction norm from three variables changes the evolutionof genetic assimilation (Ergon & Ergon, 2017). Onesolution to this dilemma is to study more complex modelswith flexible, emergent functions linking cues to pheno-types (e.g., Draghi & Whitlock, 2012). As deployed here,the core concept that the two‐parameter model isintended to capture is constraint via pleiotropic correla-tions between the expression of the same trait acrossdifferent environments.

Evaluating the role of spatial structure is a keymotivation for this model, and its importance is evidentin the role of secondarily evolved Environment‐1

specialists in the evolution of plasticity. This particularpathway to plastic generalism seems to benefit fromthe existence of a margin of underexploited patches ofthe ancestral environment, surrounded by clusters of thenovel environment. One potential follow‐up couldexamine how autocorrelation in the placement of patchesof the two environments shapes the range margin, andtherefore affects the evolution of plasticity. A moredetailed examination of the behavior of the population atits margin would also benefit by exploring the dispersalparameter as well as variants of the model that allowedfor long‐tailed dispersal kernels. A second ecologicaldimension deserving of more investigation is the demo-graphic effects of environmental noise. As seen inFigure 2, higher noise leads to a lower populationdensity, which may partially explain why newly evolvedspecialists tend to arise and succeed near the southernmargin of the landscape (Figure 3). However, higherdensities also equate to greater competition, and a futurestudy could clarify the role of these demographicconsiderations in the evolution of specialists and general-ists. While previous models have considered the impactof hard selection on the evolution of specialists andgeneralists (e.g., Van Tienderen, 1991), studies of anexplicit landscape could lead to new insights about theinteractions of space and demography.

The question of developmental biases in evolutionrelates to the issue of the evolution of phenotypicplasticity both directly and conceptually. On the mostbasic level, plasticity represents a sensitivity of thedevelopmental processes to the environment; the evolu-tion of plasticity requires a change in how information isprocessed during development that must bias howmutations can affect traits. Understanding plasticity is,therefore, one avenue toward a larger comprehension ofhow genotype‐phenotype maps are shaped by evolution,and in turn, direct its course by biasing the spectrum ofmutational effects on phenotypes. One way to appreciatethis bias is to examine the role of pleiotropy in thismodel: regardless of the genotype, any mutation in theslope or intercept parameter is clearly pleiotropic.However, the degree of constraint imposed by thispleiotropy changes with the specific, quantitativenature of that pleiotropy, with plastic Environment‐2specialists able to find mutations that are pleiotropic butstill adaptive.

Conceptually, plasticity is often viewed as an idealsolution to the problems of heterogeneity across environ-ments, and studies focus on the constraints that preventnature from realizing this ideal. This conceptual framingmirrors that of developmental variability and bias, inwhich pleiotropy and constraints are viewed as devia-tions, requiring explanations, from an ideal, isotropic

10 of 12 | DRAGHI

distribution of mutational effects (Gould, 2002). Eachviewpoint references an impossible ideal which perhapslimits our ability to see that constraints and biases areintrinsic to developmental systems, and not pathologiesin need of comparison to a version of biology withoutdevelopment. An alternative is to study how variabilityand plasticity arise out of the assembly of geneticelements to answer adaptive challenges. Here I addressboth how constraints can emerge from development andhow selection in a complex landscape can find waysaround those constraints, showing how constraints canbe relevant without being absolute.

ACKNOWLEDGMENTS

The author thanks Holly Kindsvater and the editors ofthis special issue.

CONFLICT OF INTERESTS

The authors declare that there are no conflict of interests.

ORCID

Jeremy Draghi http://orcid.org/0000-0002-7609-7836

REFERENCES

Ashander, J., Chevin, L. M., & Baskett, M. L. (2016). Predictingevolutionary rescue via evolving plasticity in stochastic environ-ments. Proceedings of the Royal Society B, 283(1839), 20161690.

Auld, J. R., Agrawal, A. A., & Relyea, R. A. (2009). Re‐evaluatingthe costs and limits of adaptive phenotypic plasticity. Proceed-ings of the Royal Society B, 277(1681), 503–511.

Bradshaw, A. D. (1965). Evolutionary significance of phenotypicplasticity in plants, Advances in Genetics (13, pp. 115–155.Academic Press.

Bruggeman, F. J., & Teusink, B. (2018). Living with noise: On thepropagation of noise from molecules to phenotype and fitness.Current Opinion in Systems Biology, 8, 144–150.

Bull, J. J. (1987). Evolution of phenotypic variance. Evolution, 41,2303–2315.

Van Buskirk, J., & Steiner, U. K. (2009). The fitness costs ofdevelopmental canalization and plasticity. Journal of Evolu-tionary Biology, 22(4), 852–860.

Bódi, Z., Farkas, Z., Nevozhay, D., Kalapis, D., Lázár, V., Csörgő, B., …Araújo, H. (2017). Phenotypic heterogeneity promotes adaptiveevolution. PLOS Biology, 15(5), e2000644.

Davidson, A. M., Jennions, M., & Nicotra, A. B. (2011). Do invasivespecies show higher phenotypic plasticity than native species and, ifso, is it adaptive? A meta‐analysis. Ecology Letters, 14(4), 419–431.

DeWitt, T. J., Sih, A., & Wilson, D. S. (1998). Costs and limits ofphenotypic plasticity. Trends in Ecology and Evolution, 13(2), 77–81.

Donaldson‐Matasci, M. C., Bergstrom, C. T., & Lachmann, M.(2013). When unreliable cues are good enough. The AmericanNaturalist, 182(3), 313–327.

Draghi, J. (2018). Links between evolutionary processes andphenotypic robustness in microbes. Seminars in Cell andDevelopmental Biology, 88, 46–53.

Draghi, J. A., & Whitlock, M. C. (2012). Phenotypic plasticityfacilitates mutational variance, genetic variance, and evolva-bility along the major axis of environmental variation. Evolu-tion, 66(9), 2891–2902.

Ergon, T., & Ergon, R. (2017). When three traits make a line:Evolution of phenotypic plasticity and genetic assimilationthrough linear reaction norms in stochastic environments.Journal of Evolutionary Biology, 30(3), 486–500.

Frank, S. A., & Slatkin, M. (1990). Evolution in a variableenvironment. American Naturalist, 136(2), 244–260.

Gould, S. J. (2002). The structure of evolutionary theory. HarvardUniversity Press.

De Jong, G., & Gavrilets, S. (2000). Maintenance of genetic variationin phenotypic plasticity: The role of environmental variation.Genetics Research, 76(3), 295–304.

Leimar, O., Hammerstein, P., & Van Dooren, T. J. (2006). A newperspective on developmental plasticity and the principles ofadaptive morph determination. The American Naturalist,167(3), 367–376.

Maxwell, C. S., & Magwene, P. M. (2017). When sensing isgambling: An experimental system reveals how plasticity cangenerate tunable bet‐hedging strategies. Evolution, 71(4),859–871.

Moczek, A. P., Sultan, S., Foster, S., Ledón‐Rettig, C., Dworkin, I.,Nijhout, H. F., … Pfennig, D. W. (2011). The role ofdevelopmental plasticity in evolutionary innovation. Proceed-ings of the Royal Society B, 278(1719), 2705–2713.

Murren, C. J., Auld, J. R., Callahan, H., Ghalambor, C. K.,Handelsman, C. A., Heskel, M. A., … Schlichting, C. D. (2015).Constraints on the evolution of phenotypic plasticity: Limitsand costs of phenotype and plasticity. Heredity, 115(4), 293–301.

Palacio‐López, K., Beckage, B., Scheiner, S., & Molofsky, J. (2015).The ubiquity of phenotypic plasticity in plants: A synthesis.Ecology and Evolution, 5(16), 3389–3400.

Pfennig, D. W., Wund, M. A., Snell‐Rood, E. C., Cruickshank, T.,Schlichting, C. D., & Moczek, A. P. (2010). Phenotypicplasticity’s impacts on diversification and speciation. Trends inEcology and Evolution, 25(8), 459–467.

Raser, J. M., & O’shea, E. K. (2005). Noise in gene expression: Origins,consequences, and control. Science, 309(5743), 2010–2013.

Scheiner, S. M., & Holt, R. D. (2012). The genetics of phenotypicplasticity. X. Variation versus uncertainty. Ecology andEvolution, 2(4), 751–767.

Scheiner, S. M. (2013). The genetics of phenotypic plasticity. XII.Temporal and spatial heterogeneity. Ecology and Evolution,3(13), 4596–4609.

Scheiner, S. M. (2014a). Bet‐hedging as a complex interactionamong developmental instability, environmental heterogeneity,dispersal, and life‐history strategy. Ecology and Evolution, 4(4),505–515.

Scheiner, S. M. (2014b). The genetics of phenotypic plasticity. XIII.Interactions with developmental instability. Ecology and Evolu-tion, 4(8), 1347–1360.

Scheiner, S. M., Caplan, R. L., & Lyman, R. F. (1991). The geneticsof phenotypic plasticity. III. Genetic correlations and fluctuatingasymmetries. Journal of Evolutionary Biology, 4(1), 51–68.

DRAGHI | 11 of 12

Schneider, R. F., & Meyer, A. (2017). How plasticity, geneticassimilation and cryptic genetic variation may contribute toadaptive radiations. Molecular Ecology, 26(1), 330–350.

Simons, A. M. (2014). Playing smart vs. playing safe: The jointexpression of phenotypic plasticity and potential bet hedgingacross and within thermal environments. Journal of Evolu-tionary Biology, 27(6), 1047–1056.

Starrfelt, J., & Kokko, H. (2012). Bet‐hedging—A triple trade‐offbetween means, variances and correlations. Biol. Rev, 87(3),742–755.

Svardal, H., Rueffler, C., & Hermisson, J. (2011). Comparingenvironmental and genetic variance as adaptive response tofluctuating selection. Evolution, 65(9), 2492–2513.

Van Tienderen, P. H. (1991). Evolution of generalists and specialistsin spatially heterogeneous environments. Evolution, 45(6),1317–1331.

Tonsor, S. J., Elnaccash, T. W., & Scheiner, S. M. (2013).Developmental instability is genetically correlated with pheno-typic plasticity, constraining heritability, and fitness. Evolution,67(10), 2923–2935.

Tufto, J. (2015). Genetic evolution, plasticity, and bet‐hedging asadaptive responses to temporally autocorrelated fluctuating

selection: A quantitative genetic model. Evolution, 69(8),2034–2049.

Veening, J. W., Smits, W. K., & Kuipers, O. P. (2008). Bistability,epigenetics, and bet‐hedging in bacteria. Annual Review ofMicrobiology, 62, 193–210.

Via, S. (1987). Genetic constraints on the evolution of phenotypicplasticity. In Loeschcke, V. (Ed.), Genetic constraints on adaptiveevolution (pp. 47–71). Berlin, Heidelberg: Springer.

Wang, Z., & Zhang, J. (2011). Impact of gene expression noise onorganismal fitness and the efficacy of natural selection. Proceedingsof the National Academy of Sciences, 108(16), E67–E76.

Wolf, L., Silander, O. K., & van Nimwegen, E. (2015). Expression noisefacilitates the evolution of gene regulation. eLife, 4, e05856.

How to cite this article: Draghi J. DevelopmentalNoise and Ecological Opportunity Across SpaceCan Release Constraints on the Evolution ofPlasticity. Evolution & Development. 2019;e12305.https://doi.org/10.1111/ede.12305

12 of 12 | DRAGHI